Author丨Juny
Editor丨Sea waist
Source丨OpenAI Spring Conference
Ten years ago, a film called "Her" won the 86th Academy Award for Best Original Screenplay in 2014, and it tells the story of a lonely writer who falls in love with an artificial intelligence voice assistant on his phone. In the movie, this AI named Samantha has a hoarse and sexy voice, she is humorous, empathetic, and accompanies the male protagonist anytime and anywhere, and gradually becomes an indispensable part of his life.
And now, 10 years later, at the spring conference held by OpenAI, with the arrival of the new GPT-4o model, Samantha has officially become a reality. The upgraded version of ChatGPT can not only chat with you naturally like Samantha, but even observe and understand your emotions through your phone's camera.
In fact, in the past year, the business of "AI accompaniment" has always been a key area for artificial intelligence companies to compete for layout, from the perspective of business realization, "AI accompaniment" is currently one of the application scenarios that consumers are most willing to pay, and even a large group of foreigners on TikTok have already been collectively "harvested" by various AI escort products. This time, OpenAI's model update and large-scale open trial have obviously pushed this already hot track to a new dimension.
1. The era of your AI friend has arrived
The OpenAI launch event was very brief, only 26 minutes long, but the evolution of ChatGPT was enough to be breathtaking.
Although GPT-5 did not arrive as scheduled, OpenAI's latest flagship model, GPT-4o, has undergone a "qualitative change" in human-computer interaction. According to the official introduction, the "o" in 4o is the abbreviation of "omni", which means that this version of GPT has opened up all the capabilities of text, vision, audio, etc., and can accept any combination of input and output. With a minimum of 232 milliseconds and an average of 320 milliseconds for audio input response time, it has now reached the speed of a human response in a conversation.
With the support of GPT-4o, how natural can ChatGPT's voice feedback be? During the live demonstration, OpenAI's staff asked ChatGPT how it could ease his nervousness. Then, our "Samantha" appeared, a gentle and natural female voice suggested that he take a deep breath, and when he heard the other person breathing loudly, he immediately joked with him that you are not a vacuum cleaner, and when he heard the other person exhale and exhale steadily, he immediately encouraged and praised.
Not only can it give you daily advice and support, but today's ChatGPT has been completely upgraded to become an all-round "friend". For example, she can tell you bedtime stories in a variety of tones and moods, and her voice can be high-pitched, low-pitched, or even sung to you in the form of a musical.
It can also be your simultaneous translator, helping you switch between different languages silkily in any scenario.
In addition, with the blessing of visual ability, ChatGPT can not only tutor you with homework, problems, and code online.
You can also take a closer look at your appearance and emotions through the camera, and in the process, it will act like a friend, asking questions, laughing and giving you advice.
OpenAI's technical report shows that GPT-4o has made particularly significant improvements in visual and audio understanding, not only to interrupt conversations at any time, but also to automatically change pitch and emotions in a variety of different scenarios. In terms of language, knowledge, mathematics, and programming, it surpasses competitors such as GPT-4T, Claude 3 Opus, and Gemini. Developers can now also access GPT-4o as a text and visual model in the API. GPT-4o is 2x faster, half the price, and rate limiting 5x faster than GPT-4 Turbo.
OpenAI said that the voice conversation feature will be first available to paying users of ChatGPT in the coming weeks, and free users can also experience the text and image features powered by GPT-4o in ChatGPT starting today.
2. The "battleground" of AI companies
OpenAI's update today seems to have brought the previously criticized AI voice assistant back to the center of the stage. But in fact, with the maturity of large models, in the past year, a large number of AI companies have laid out in advance in this track and even carried out a series of commercialization and monetization attempts. It's just that their products do not appear in the form of traditional voice assistants in mobile phones and devices, but are wrapped in the concept of "AI companionship".
Nowadays, on TikTok, when you type in keywords like "AI dating" and "AI companion", you will find that a large number of related products and recommended views on the platform are above the million level. Some of them are in the form of two-dimensional, cartoon images combined with AI, and some are real-life AI images. Among them, the most out-of-the-circle at present include Character.ai, CrushOn, Talkie, Replika and so on.
Source: TikTok
Compared with ChatGPT, which focuses more on functional attributes, these AI products focus more on emotional companionship and emotional value, with the goal of providing users with a personalized social experience in a language that is closer to real people. Judging from the current effect, the user stickiness of AI companion products is much higher than that of functional AI products. Functional AI products are often used in situations where specific needs and solutions are sought, but the time and effort that people put into interacting with AI-accompanied products is transformed into an emotional sustenance, making it a long-term interaction bond.
For example, the Character.ai, founded by former Google engineers, has more than 20 million registered users and is valued at more than $5 billion since its launch in September 2022. According to Similarweb's statistics, Character.ai website visits have maintained steady growth over the past year, with a cumulative number of visits of more than 620 million in the past three months, and Character.ai is expected to generate more than $16 million in revenue in 2024 while most AI companies are still burning cash.
Talkie has also recently made a splash on this track. Talkie is a product of MiniMax, an AI startup founded by Yan Junjie, former vice president of SenseTime Technology and head of general intelligent technology, and has seen explosive growth since its official launch overseas in June last year. Since the beginning of this year, the growth rate of visits to its web terminal has increased by a thousand times, and the overall popularity is second only to Character AI in AI companion applications.
图源：similarweb
Similar to Character AI, Talkie is also aiming at the AI chat market. However, in terms of gameplay, it not only provides more personalized settings in terms of AI image, character design, and voice, but also combines more elements such as card games and game plots, so that users have a stronger sense of immersion and entertainment in the process of interacting with AI, and the monetization channels are also more diversified.
Last year, A16Z pointed out that AI companions will be the first killer applications of artificial intelligence, which can truly bring generative AI into consumers' daily lives. In fact, if we pay a little attention, we will also find that at present, both startups and technology giants are almost quietly laying out the track of "AI accompaniment". For example, Inflection AI's emotional chatbot Pi, Snapchat's full opening of "May AI", and Meta's AI character chatbot based on its own model. In China, Tencent's "Unaccompanied", Baidu's "Xiaokan Planet", and Meituan's "Wow" are all aiming at the AI virtual chatbot track, and some leading large-scale model startups are also trying to take Talkie's overseas route recently, such as Mona Land based on Zero One Thousand Things, Joyland based on West Lake Chenxin, and it is reported that the dark side of the moon is also working on an AI chat product called Ohai AI recently.
In addition, it is widely speculated that at Google's IO conference tomorrow, its latest artificial intelligence model Gemini will once again be improved in terms of language and emotional capabilities, similar to ChatGPT this time, making AI closer to "real people".
3. Function + emotion, from "AI accompaniment" to "system assistant"
Before today's OpenAI update, AI assistants, AI companions, and AI tools all seem to be scattered in different fields. ChatGPT, Copilot and the like are helpers at work, while Character.ai and Talkie are emotional sustenance and pastime in their leisure time. But one trend that is evident in today's conference is that functional AI and emotional AI are converging.
AI chat and companionship capture people's unconstrained and more authentic social and emotional needs in the virtual social experience, and users can get a "friend" who is always meticulous and patient, lively and interesting and sincere enough to you by choosing the AI character they like. At the same time, AI's task-solving capabilities can help users solve specific problems and become an "assistant" that provides suggestions and solutions. The combination of function + emotion not only makes the concept of "AI companionship" richer, but also drives the role positioning of AI assistants to fundamentally change in people's lives.
With the in-depth development of multimodality, AI can not only write and speak, but also see and think everything that humans can perceive in the future, and meet people's social, emotional, companionship, support and other needs in an all-round way, and gradually become an indispensable part of people's lives.
GPT-4o's multimodal capabilities, image courtesy of OpenAI
Nowadays, AI companies are seizing the market of "AI accompaniment", and a very important significance is to take the lead in gaining some users, starting from the emotional dimension and then expanding to systematic AI assistants. Perhaps soon, the gateway for people to interact with the world will become AI, and talking to AI friends on mobile phones will replace almost all the actions we need to take on a daily basis.
Although this is only a small update for OpenAI on the surface, perhaps just like the name of the latest GPT model "Omni", its all-encompassing capabilities will soon leverage the evolution of the entire system.